https://www.irjet.net/archives/V9/i4/IRJET-V9I4297.pdf

1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 04 | Apr 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2566
Dog Breed Prediction System (Web)
Gaurav Acharya1, Rithik Borade2, Piyush Shejwal3, Prof. Suchetadevi Gaikwad4
1,2,3Dept of Information Technology Engineering, Atharva College of Engineering
4Professor, Dept of Information Technology Engineering, Atharva College of Engineering, Maharashtra,
India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract -Deep learning allows us to train algorithms to
handle picture classification challenges in a healthy way. The
Convolutional Neural Network (CNN) is a widely used image
classification and detection technology. We define a CNN-
based method for identifying dog breedsperclassimagein this
paper. Stanford University provided the data set for this
investigation. Thesystemextractscharacteristicsusingseveral
models and then combines them to construct a deep net that
predicts dog breed based on the greatest probabilityobtained.
The algorithm CNN delivers the most accurate and efficient
prediction, according to an experimental technique and
comparative analysis of different methodologies.
The system predicts with changes to the epoch value, weights,
and other data augmentation strategies.
Key Words: Machine Learning, Tensorflow,
Classification, CNN, Deep Learning
1. INTRODUCTION
Among pets, dogs are the most popular animals. Customers
interested in purchasing a dog can conduct research by
observing canines owned by others. It may not always be
easy to approach the dog's owner and inquire about the
breed. Additionally, it's possible that thedog'sownersaren't
aware of their dog's breed. We attempted to solve this
market need in this paper by automating the process of
locating a dog breed via an Android application that allows
you to determine the breed of a dog just by snapping a
picture of it. CNNs (Convolutional Neural Networks) mimic
human vision. In simple terms, we can use our vision to
categorise thing. CNN develops software that enablesvisual
identification and recognition by constructing a
mathematical model and putting it into practice as an
algorithm. CNNs are a subset of deep learning and feed
forward artificial neural networks. It functions in the same
way as ordinary neural networks,exceptthateveryinput isa
picture. As a result of the assumption, we can hide a few
properties from the architecture, making the feed forward
function more efficient. CNNs have been shown to be useful
at analysing visual data. They use multi-layer perceptrons
and, in comparison to other image classification algorithms,
usually require less preparation.
1.2. OBJECTIVES
Creating a system to predict the breed of the dog correctly
with the highest possible accuracy. To perform and build a
user interface so as to make a system that makes the Dog
breed prediction easy through applications of ML. To
implement this project using the machine/deep learning
algorithms. To showtheinformationaboutthepredicted dog
2. LITERATURE SURVEY
Different papers and articles have been reviewed for this
project. Also, their conclusions are summarized in this
section. The section present documents that were studied
prior and post project development. The mentioned articles
provide with a better understanding about structure of the
system and how various algorithms could be combined
together so as to build a system with higher efficiency.
Table -1: Publications Cited:
Title Year Author Summary
Learning
Features and
Parts forFine-
Grained
Recognition22
nd Int.
Conference on
Pattern
Recognition
2015
IRJET
Jonatn
Krause;Tim
nit
Gebru;Jia
Deng;Li-Jia
Li
Just an overviewof
learning how to
train different
Models. Datasets
include collection
of their own
pictures.
Ensemble deep
neural
networks for
domain-specific
Image
Recognition
2016
IEEE
(ICME
W)
Wenbo
Li;ChuanKe
Limited Images
were taken into
consideration for
training. Datasets
included were not
mentioned
Dog Breed
Identification
UsingDeep
Learning
2018I
EEE
Zalán
Ráduly;Csab
a Sulyok
The dataset which
wastaken in
consideration is
very less, Resulting
in lower accuracy
ResNet-82.01%
Transfer
Learning on
Convolutional
Neural
Networksfor
Dog
Identification
2018I
EEE
Xinyuan
Tu;Kenneth
Lai
The number ofdog
breedtaken for
training was
limited to 100
images perclass.
The model used
and accuracy
gained was
SFU/Resnet-

2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 04 | Apr 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2567
92.34%
Dog
Identification
using Soft
Biometrics and
Neural
2019
IEEE
Xinyuan
Tu;Svetlana
Yanushkevi
ch
Biometrics used in
this project needs
device of much
higher cost to be
involved
SFU/DenseNet-
90.22%
Transfer
Learning
Approach to
Fine Grained
Image
Classification
2019 Int.
Russian
Automation
Conference
2019
IEEE
ValentinA.
Golodov;Ma
riya S.
Dubrovina;
Anast asiya
S. Paziy
Limitedto
European dog
breeds. The model
used and accuracy
gained was
SFU/VGG16
-91.78%
Identification
of Dogs breed
for deep
learning
2019
IEEE
(IACC)
Rakesh
Kumar;Man
ish
Sharma;Kri
tika
Dhawale;Ga
urav Singal
Used lesser Epoch
value augmentat
ion. The model
used and accuracy
gained was
SFU/Inception
model- 85.67%
Dog-Breed-
classifier for
facial
recognition by
using CNN
2020 Bickey
Kumar
shah;Aman
Kumar;Amr
it Kumar
Only the Images of
Dog face would be
taken into
considerat-
ion. The
model used and
accuracy gained
was SFU/Mobile
Net-75.2%
3. TECHNICAL DEFINATION
3.1 DEEP LEARNING
Deep learning is a type of machine learning method that
employs numerous layers to extract higher-level features
from unprocessed data. Lowerlayersinimageprocessing,for
example, may recognise edges, whereas higher layers may
identify human-relevant notions like numerals, letters, or
faces.
3.2 CONVOLUTIONAL NEURAL NETWORK
CNN (Convolutional Neural Network) is an acronym for
Convolutional Neural Network. Convolutional Neural
Networks (CNNs) are a sort of Deep Learning and a unique
type of progressing artificialneural network that isactivated
by the visual brain. The neuron in a layer in CNN is only
related to a small area of the layer before it, rather than all
the neurons in a completelyrelatedmanner,soCNNmanages
fewer weights and a much smaller number of neurons.
3.3 ResNet
The structure of this network is built in such way that the
convolutional neural network would run in an efficient
manner. ResNet is a powerful backbone model that is used
very frequently in many computer vision tasks. ResNet uses
skip connection to add the output from an earlier layer to a
later layer. This helps it mitigate the vanishing gradient
problem making it possible to constructnetworkswithupto
thousands of convolutional layers, which outperform
shallower networks. However, the adding some of deep
layers to a network regularly can affect the degradation of
the result. It is known as problem of vanishing gradient
where in neural networks, while getting skilled via lower
backpropagation, relies on the gradient descent, lowering
the loss feature to find the lowering weights.
3.4 INCEPTION
In a CNN,an InceptionModuleisanimagemodelblockwhose
goal is to approximatean ideal local sparse structure. Simply
said, it lets us to employ many sorts of filter sizes in a single
image block, rather than being limited to a single filter size,
which we then concatenate and pass onto the next layer.
4. PROPOSED SOLUTION
A Dog breed prediction system is built using the
convolutional neural network (CNN). The system can be
useful forthe NGO’s whocomeacrossalotofabandoneddogs
by knowing about the breeds of the dog. Having the
knowledge of dogs breed would prove to be a great asset for
the NGO’s for treating the dog accordingly. Using the
Streamlit module, the system would be more more
interactive for the user.
Using Streamlit, a dynamic system could be built for the user
convenience. It can be usedbytheusersfortheirconvenience
to predict different dog breed.
Around 120 breeds of dog from different parts of the world
would be taken into consideration. For training the system a
large dataset would be taken into considerationwhichwould
be divided into training and testing purpose.
4.1 REQUIREMENTS
Anaconda
Anaconda Navigator is a GUI tool that is included in the
Anaconda distribution.Itallowsusertocarryoutexecutionof
complex projects efficiently. It offers important platformlike
Jupyter Notebook. Anaconda also provides a command
prompt.
Google Colaboratory
Google Colab is a product from Google Research. It is most
suitable for machine learning system. Google Colab provides
an in-built GUI so as to carry out any User Interface related
problems efficiently.

3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 04 | Apr 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2568
Technologies used:
Python
Python is an interpreted high-level general-purpose
programming language. Python is dynamically-typed and
garbage-collected. Python is preferred for machine learning
as it is easy language to code. It also supports essential
libraries for conducting the im age classification process.
Tensorflow
TensorFlow is one the important libraries available in
python. It is a free and open-source software library for
machine learning and AI. It is more oftenly used for carrying
out problems using deep neural networks.
Streamlit
Streamlit is an open-source Python library. It is mainly used
as for ML. It provides with a dynamic approach towards the
sytem.
4.2. WORKFLOW OF THE SYSTEM
1. Data Flow Diagram
In corporate statistics, data flow diagrams are used to
graphically represent the data flow. DFD denotes the
methods used in a model to transfer data from the front to
the record garage and generate reports. Logical andphysical
data float diagrams can be separated. The logical data
diagram is a diagram that depicts information using a model
to carry out an organization's positive capabilities.
Fig.1- Data Flow Diagram
4.3. Data Collection and Preprocessing
Data series is described because the system of collecting,
measuring, and reading correctinsightsforstudiestheusage
of fashionable proven techniques. A researchercancompare
their speculation on the premise of amassed facts. In
maximum cases, facts series is the number one and
maximum crucial step for studies,regardlessofthesphere of
studies. The technique of facts series is extraordinary for
extraordinary fields of study, relying on the specified
information.
4.4 User Interface
The interface is mainly divided into four section
Fig.2 - Interface displayed to the user
Fig 3 – Displaying the input image
Fig.4 - System predicting dog’s breed correctly
5 METHODOLOGY
a) In the first phase, we select an image from which we wish
to forecast an object. Here, we select a dog image for system.
b) We execute object recognition in the second phase, which
will connect to Tensorflow. And which contains around
10,000 photos of various dog breeds for training purpose.

4. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 04 | Apr 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2569
c) The CNN (convolution Neural Network) model, which has
been pre-trained, is loaded in the third phase. For
comparison, a CNN with two convolutional layers selects a
unique combination of activation characteristics and
classifiers. Training datasetsareusedtotestthesysteminthe
fourth phase.
d) The input image is then passed as an input parameter to
ReadImage(), which turns the image to pixels in the fifth
phase.
e) Following that, Input() is called, which turns the pixel
picture to a tensor.
f) Finally, this tensor picture is created and this tensor image
is passedasan inputparametertotheclassifyfunction,which
predicts the output based on probability, with the highest
probability value being considered the best predicted breed.
6 CONCLUSIONS AND FUTURE SCOPE
Dog breed prediction using deep learning is developed using
convolutional neural network to predict the breed of dogsby
taking their images as input. This system is easy and
convenient for users to predict the dog’s breed. The system
would predict the dog breeds at a much higher accuracy. We
have demonstrated modified approach of the state of art
network.
As further work, due to increasing the size of datasets using
some optimization methods to find the best hyper-
parameters for learning rate in order to reduce the
computational time of test set besides the classification
accuracy in the test set. Varying epochs can also reduce the
timecomplexityand provideopportunitytohavehigherlevel
of augmenting in dataset to increase the accuracy of
classification on the datasets.
Fig. 5 – Comparative Analysis
7. REFRENCES
1. Keith D.Foote, (2017). A Brief History of Deep
Learning Available:
https://www.dataversity.net/briefhistory- deep-
learning/
2. Arora, A.,Candel, A., Lanford, J., LeDell, E. andParmar,
V. (2015). Deep Learning with H2O. 3rd ed.
[ebook]Availableat:https://h2orelease.s3.amazonaws.
com/h2o/master/3190/docswebsite/h2odocs/booklet
s/ DeepLearning_Vignette.pdf
3. Nokwon Jeong, Soosun Cho (2017) „Instagram image
classification with Deep Learning‟.
4. J. Salamon and J. P. Bello, Deep Convolutional Neural
Networks and Data Augmentation for Environmental
Sound Classification, no. November, pp. 15,2016.5.H.C.
Shin, L. Lu, L. Kim, A. Seff, J. Yao, and R. M. Sum- mers.
Interleaved text/image deep mining on a very large-
scale radiology database.
5. A. Ahmeda, H. Yousifa, R. Kaysb and Z. Hea, "Semantic
region of interest and species classification in the deep
neural network featuredomain",EcologicalInform.,vol.
52, pp. 57-68, Jul. 2019.